On Fri, 18 Nov 2011 08:53:00 -0800
Jesse Brandeburg <[email protected]> wrote:

> On Thu, 17 Nov 2011 17:46:46 -0800
> Flavio Leitner <[email protected]> wrote:
> 
> > On Fri, 18 Nov 2011 09:37:12 +0800
> > Michael Wang <[email protected]> wrote:
> > 
> > > From: Michael Wang <[email protected]>
> > > 
> > > Descriptors may not be write-back while checking TX hang with flag
> > > FLAG2_DMA_BURST on.
> > > So when we detect hang, we just flush the descriptor and detect
> > > again for once.
> > > 
> > > Signed-off-by: Michael Wang <[email protected]>
> > 
> > Jesse,
> > This is tested and fixes the issue I've reported in the other
> > thread: [E1000-devel] 82571EB: Detected Hardware Unit Hang
> > http://sourceforge.net/mailarchive/forum.php?thread_name=20111014140426.3d576173%40asterix.rh&forum_name=e1000-devel
> > 
> > Signed-off-by: Flavio Leitner <[email protected]>
> 
> Flavio/Michael, thanks for working on this, the patch itself seems
> okay, but it does increase the time to detect a tx hang doesn't it?

Yes, that is correct. The specific models having the flag
FLAG2_DMA_BURST will wait another watchdog round for the
detection.  We thought about scheduling the watchdog in a
short time to reduce the impact, but we don't know how much
time it takes to finish up the write-back. There is the 
interruption but then the fix would be rather large and
complex, I think.

> I'm okay with the patch functionality because you're implementing
> (effectively, if not a little indirectly) the fix our hardware
> engineer suggested which was two writes to the FPD bit.

Yes, we found about FPD register then I did a patch to create
a workqueue periodically writing FDP register. It had fixed
the issue as well. It was more like an experiment to confirm
the root cause without touching the driver's work flow.

The proposed patch avoids writing during normal situation to
keep the performance up. Otherwise, if we wrote to FPD each
time before watchdog, we could have up to 4 descriptors being
transferred with no good reason.

> We can test the patch in our lab here, Jeff Kirsher will push it
> upstream when it completes testing.

Ok, sounds like a plan to me.
Thanks for reviewing it.
fbl

------------------------------------------------------------------------------
All the data continuously generated in your IT infrastructure 
contains a definitive record of customers, application performance, 
security threats, fraudulent activity, and more. Splunk takes this 
data and makes sense of it. IT sense. And common sense.
http://p.sf.net/sfu/splunk-novd2d
_______________________________________________
E1000-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/e1000-devel
To learn more about Intel&#174; Ethernet, visit 
http://communities.intel.com/community/wired

Reply via email to